In this paper, we present a method that learns a deterministic finite state machine
from the conformance test logs of a telecommunication protocol; then that machine
is used as test model for performance testing. The learning process is in contrast
to most theoretical methods automatic; it applies a sequential pattern mining algorithm
on the test logs, and uses a recently proposed metric for finding frequent and significant
transition sequences. The method aims to help and speed up test model design, and
at the same time it may not provide an exact solution, the equivalence of some states
may not be proven. In the paper, we show the results of experiments on random machines,
and issues and considerations that arise when the method was applied to 3GGP Telephony
Application Server test logs.
